Comparison of Sentiment Tools across Domains

Hint: Try executing this chunk by clicking the Run button within the chunk or by placing your cursor inside it and pressing Cmd+Shift+Enter.

# load required libraries

# to use harry potter dataset
# devtools::install_github("bradleyboehmke/harrypotter")
# devtools::install_github("quanteda/quanteda.sentiment")
# devtools::install_github("quanteda/quanteda.corpora")

library(quanteda)
library(readtext)
library(corpus)
library(tidyverse)
library(stringr)
library(tidytext)
library(harrypotter)
library(dplyr)
library(quanteda.sentiment)
library(vader)


require(quanteda)
require(quanteda.corpora)
require(quanteda.sentiment)
#library("quanteda", warn.conflicts = FALSE, quietly = TRUE)

afinn2 <- data_dictionary_AFINN

Harry Potter - Dataset

# load harry potter dataset 
titles <- c("Philosopher's Stone", "Chamber of Secrets", "Prisoner of Azkaban",
            "Goblet of Fire", "Order of the Phoenix", "Half-Blood Prince",
            "Deathly Hallows")

books <- list(philosophers_stone, chamber_of_secrets, prisoner_of_azkaban,
           goblet_of_fire, order_of_the_phoenix, half_blood_prince,
           deathly_hallows)
  
series <- tibble()

for(i in seq_along(titles)) {
        
        clean <- tibble(chapter = seq_along(books[[i]]),
                        text = books[[i]]) %>%
             #unnest_tokens(word, text) %>%
             mutate(book = titles[i]) %>%
             select(book, everything())

        series <- rbind(series, clean)
}

series$book <- factor(series$book, levels = rev(titles))

series

#book_groups <- series %>% group_by(book, chapter)
# tokenize hp1
#hp1_tokenized <- tokens_tolower(tokens(philosophers_stone, remove_punct = TRUE)) 

Harry Potter - AFINN Lexicon

Lexicoder: HP

# select only the "negative" and "positive" categories
#data_dictionary_LSD2015_pos_neg <- data_dictionary_LSD2015[1:2]
#hp1_lsd <- tokens_lookup(hp1_tokenized, dictionary = data_dictionary_LSD2015_pos_neg)

polarity(data_dictionary_LSD2015) <- 
  list(pos = c("positive", "neg_negative"), neg = c("negative", "neg_positive"))

hp1_lsd <- textstat_polarity(hp1_tokenized, data_dictionary_LSD2015)

hp1_lsd_tokens <- tokens_lookup(hp1_tokenized, data_dictionary_LSD2015, nested_scope = "dictionary", exclusive = FALSE)
hp1_lsd.df <- as.data.frame.matrix(hp1_lsd)
hp1_lsd.df$chapter <- 1:nrow(hp1_lsd.df)

plot <- ggplot(hp1_lsd, aes(x =hp1_lsd.df$chapter, y=sentiment)) +
          geom_bar(alpha = 0.8, stat = "identity", show.legend = FALSE)
plot + ylim(-1.0, 1.0) + labs(y="sentiment", x = "chapter") + ggtitle("HP1 - Lexicoder")

AFINN: HP

hp1_afinn2 <- textstat_valence(hp1_tokenized, afinn2, normalize="dictionary")

hp1_afinn2.df <- as.data.frame.matrix(hp1_afinn2)
hp1_afinn2.df$chapter <- 1:nrow(hp1_afinn2.df)

plot <- ggplot(hp1_afinn2.df, aes(x =hp1_afinn2.df$chapter, y=sentiment)) +
          geom_bar(alpha = 0.8, stat = "identity", show.legend = FALSE)
plot + ylim(-1.0, 1.0) + labs(y="sentiment", x = "chapter") + ggtitle("HP1 - AFINN")

Warnung: Use of `hp1_afinn2.df$chapter` is discouraged. Use `chapter` instead.

VADER: HP

QUANTEDA.SENTIMENT

AFINN: HP

# Work with quanteda.sentiment on HP corpus:
# convert tibble to dataframe
series.df <- as.data.frame(series)

# tokenize books
series_tokenized <- series.df %>%
  unnest_tokens(tokens, text)

# apply afinn lexicon
series_tokenized$afinn2 <- textstat_valence(series_tokenized$tokens, afinn2)$sentiment

# replace all 0 values with na
series_tokenized[series_tokenized == 0] <- NA

series_tokenized %>%
  group_by(book, chapter) %>% # group df by book and chapter to get sentiment per chapter
  summarise(sentiment = mean(afinn2, na.rm = TRUE)) %>% # calculate mean w/o regarding na values
  mutate(method = "AFINN") %>% # add column with method 
        ggplot(aes(chapter, sentiment, fill = book)) + # plot sentiment of books
          geom_bar(alpha = 0.8, stat = "identity", show.legend = FALSE) +
          facet_wrap(~ book, ncol = 2, scales = "free_x") +
          ggtitle("AFINN HP")

`summarise()` has grouped output by 'book'. You can override using the `.groups` argument.

Lexicoder: HP

# Work with quanteda.sentiment on HP corpus:
# apply lexicoder lexicon
series$lsd <- textstat_polarity(tokens(series$text), data_dictionary_LSD2015)$sentiment 

#series.df <- as.data.frame(series)

plot <- ggplot(series, aes(chapter, lsd, fill = book)) + # plot sentiment of books
          geom_bar(alpha = 0.8, stat = "identity", show.legend = FALSE) +
          facet_wrap(~ book, ncol = 2, scales = "free_x") +
          ggtitle("Lexicoder HP")
plot 

Vader: HP

REVIEWS DATASET

# load dataset
reviews <- readtext("datasets/goodreads_reviews_children_2.json", text_field = "review_text")

# convert to dataframe
reviews.df <- as.data.frame(reviews)

# add doc_id (i.e. according to index)
reviews.df$doc_id <- 1:nrow(reviews.df)

Sample Dataset

# get random sample of 50 reviews 
reviews_sample <- reviews.df[sample(1:nrow(reviews.df), 50,
   replace=FALSE),]

# get first 50 rows of data 
reviews_50 <- head(reviews.df,50)
reviews_50 = subset(reviews_50, select = c(doc_id,text,rating))

Get Translations of Dataset

# either via corpus 
reviews.corpus <- corpus(reviews)
docvars(reviews.corpus, "language") <- textcat(reviews.corpus)
reviews_en <- corpus_subset(reviews.corpus, language == "english", drop_docid = TRUE)

# or via dataframe logic
reviews.df$language <- textcat(reviews.df$text)

Sentiment Analysis on Reviews Dataset

AFINN

# Afinn
# tokenize 
reviews_tokenized <- reviews_50 %>%
  unnest_tokens(tokens, text)

# apply afinn lexicon
reviews_tokenized$afinn2 <- textstat_valence(reviews_tokenized$tokens, afinn2)$sentiment

# replace all 0 values with na
reviews_tokenized[reviews_tokenized == 0] <- NA

# calculate mean scores for tokens per doc
afinn_scores <- reviews_tokenized %>%
  group_by(doc_id) %>% # group df by doc_id to get mean sentiment score
  summarise(total = mean(afinn2, na.rm = TRUE)) #%>% # calculate mean w/o regarding na values

# add afinn scores to df 
reviews_50$afinn <- afinn_scores$total

# different version to get plot 
reviews_tokenized %>%
  group_by(doc_id) %>% # group df by book and chapter to get sentiment per chapter
  #reviews_tokenized$sentiment = mean(afinn2, na.rm = TRUE) %>%
  summarise(sentiment = mean(afinn2, na.rm = TRUE)) %>% # calculate mean w/o regarding na values
  mutate(method = "AFINN") %>% # add column with method 
        ggplot(aes(doc_id, sentiment, fill = doc_id)) + # plot sentiment of books
          geom_bar(alpha = 0.8, stat = "identity", show.legend = FALSE) +
          #facet_wrap(~ doc_id, ncol = 2, scales = "free_x") +
          ggtitle("AFINN Reviews")

Warnung: Removed 1 rows containing missing values (position_stack).

LEXICODER

# apply lexicoder lexicon
reviews_50$lsd <- textstat_polarity(tokens(reviews_50$text), data_dictionary_LSD2015)$sentiment 
#series.df <- as.data.frame(series)

plot <- ggplot(reviews_50, aes(doc_id, lsd, fill = doc_id)) + # plot sentiment of books
          geom_bar(alpha = 0.8, stat = "identity", show.legend = FALSE) +
          #facet_wrap(~ doc_id, ncol = 2, scales = "free_x") +
          ggtitle("Lexicoder Reviews")
plot 

Vader

reviews_50$vader <- vader_df(reviews_50$text)$compound

plot <- ggplot(reviews_50, aes(doc_id, vader, fill = doc_id)) + # plot sentiment of books
          geom_bar(alpha = 0.8, stat = "identity", show.legend = FALSE) +
          #facet_wrap(~ doc_id, ncol = 2, scales = "free_x") +
          ggtitle("Vader Reviews")
plot 

Statistics

# convert to binary results
reviews_50$afinn_binary[reviews_50$afinn > 0] <- "pos"
reviews_50$afinn_binary[reviews_50$afinn <= 0] <- "neg"

reviews_50$lsd_binary[reviews_50$lsd > 0] <- "pos"
reviews_50$lsd_binary[reviews_50$lsd <= 0] <- "neg"

reviews_50$vader_binary[reviews_50$vader > 0] <- "pos"
reviews_50$vader_binary[reviews_50$vader <= 0] <- "neg"

reviews_50$rating_binary[reviews_50$rating >= 3] <- "pos"
reviews_50$rating_binary[reviews_50$rating < 3] <- "neg"

# optionally: convert 0 = negative, 1 = positive
reviews_50[reviews_50 == "pos"] <- 1
reviews_50[reviews_50 == "neg"] <- 0

actual_values <- test$rating_binary
predict_values <- test$afinn_binary

# create confusion matrix 
confusion_matrix <- table(ACTUAL=actual_values, PREDICTED=predict_values)

# assign values from matrix to true/false positives/negatives
TN <- confusion_matrix[1]
FN <- confusion_matrix[2]
FP <- confusion_matrix[3]
TP <- confusion_matrix[4]

# calculate statistics
precision <- TP/(TP+FP)
accuracy <- (TP+TN)/(TP+TN+FP+FN)
recall <- TP/(TP+FN)
F1 <- (2*precision*recall)/(precision+recall)

get_statistics <- function(df) {
  statistics <- data.frame(matrix(ncol=4, nrow=0))
  x <- c("accuracy", "precision", "recall", "F1")
  colnames(statistics) <- x
  lex1 <- "afinn_binary"
  lex2 <- "lsd_binary"
  lex3 <- "vader_binary"
  gold <- "rating_binary"
  
  lexicons <- c(lex1,lex2,lex3)
  
  for(lex in lexicons){
    confusion_matrix <- table(ACTUAL=df[[gold]], PREDICTED=df[[lex]])
    TN <- confusion_matrix[1]
    FN <- confusion_matrix[2]
    FP <- confusion_matrix[3]
    TP <- confusion_matrix[4]
  
    # calculate statistics
    precision <- TP/(TP+FP)
    accuracy <- (TP+TN)/(TP+TN+FP+FN)
    recall <- TP/(TP+FN)
    F1 <- (2*precision*recall)/(precision+recall)
  
    # add to table
    output <- c(accuracy,precision, recall, F1)
    statistics[lex,] = rbind(statistics[[lex]], output)
    }
    
  return(statistics)
  
}

get_statistics(test)